[SOL] Invert indexes for dynamic stack frames in v3 (#178)

LucasSte · web-flow · commit cc8f60e84604 · 2025-11-24T19:00:58.000-03:00
* Invert indexes for dynamic stack frames in v3

* Fix rebase conflicts
diff --git a/llvm/lib/Target/SBF/SBFFrameLowering.cpp b/llvm/lib/Target/SBF/SBFFrameLowering.cpp
@@ -21,20 +21,26 @@ bool SBFFrameLowering::hasFPImpl(const MachineFunction &MF) const { return true;
 
 void SBFFrameLowering::emitPrologue(MachineFunction &MF,
                                     MachineBasicBlock &MBB) const {
-  if (!MF.getSubtarget<SBFSubtarget>().getHasDynamicFrames()) {
+  const SBFSubtarget& Subtarget = MF.getSubtarget<SBFSubtarget>();
+  if (!Subtarget.getHasDynamicFrames()) {
     return;
   }
+
   MachineBasicBlock::iterator MBBI = MBB.begin();
   MachineFrameInfo &MFI = MF.getFrameInfo();
   int NumBytes = (int)MFI.getStackSize();
-  if ((NumBytes || MF.getSubtarget<SBFSubtarget>().getHasStaticSyscalls()) &&
-      MBBI != MBB.end()) {
+
+  if (NumBytes && MBBI != MBB.end()) {
     DebugLoc Dl = MBBI->getDebugLoc();
     const SBFInstrInfo &TII =
         *static_cast<const SBFInstrInfo *>(MF.getSubtarget().getInstrInfo());
+
+    if (Subtarget.isDynamicFramesV1())
+      NumBytes = -NumBytes;
+
     BuildMI(MBB, MBBI, Dl, TII.get(SBF::ADD_ri), SBF::R10)
         .addReg(SBF::R10)
-        .addImm(-NumBytes);
+        .addImm(NumBytes);
   }
 }
 
diff --git a/llvm/lib/Target/SBF/SBFRegisterInfo.cpp b/llvm/lib/Target/SBF/SBFRegisterInfo.cpp
@@ -167,13 +167,20 @@ int SBFRegisterInfo::resolveInternalFrameIndex(
 
   if (SubTarget.getHasDynamicFrames() &&
       SBFFuncInfo->containsFrameIndex(FI)) {
-    return -Offset;
+    if (SubTarget.isDynamicFramesV1())
+      return -Offset;
+
+    return Offset;
   }
 
   Offset += Imm.value_or(0);
 
   if (SubTarget.getHasDynamicFrames()) {
-    return static_cast<int>(StackSize) + Offset;
+    Offset += static_cast<int>(StackSize);
+    if (SubTarget.isDynamicFramesV1())
+      return Offset;
+
+    return -Offset;
   }
 
   return Offset;
diff --git a/llvm/lib/Target/SBF/SBFSubtarget.cpp b/llvm/lib/Target/SBF/SBFSubtarget.cpp
@@ -57,6 +57,7 @@ void SBFSubtarget::initializeEnvironment(const Triple &TT) {
   HasStaticSyscalls = false;
   IsAbiV2 = false;
   HasJmp32 = false;
+  HasDynamicFramesV3 = false;
 }
 
 void SBFSubtarget::initSubtargetFeatures(StringRef CPU, StringRef FS) {
diff --git a/llvm/lib/Target/SBF/SBFSubtarget.h b/llvm/lib/Target/SBF/SBFSubtarget.h
@@ -92,6 +92,9 @@ class SBFSubtarget : public SBFGenSubtargetInfo {
   // JMP32 support depends on ALU32 being enabled
   bool HasJmp32;
 
+  // Whether we are dealing with dynamic stack frames in SBPFv3
+  bool HasDynamicFramesV3;
+
   std::unique_ptr<CallLowering> CallLoweringInfo;
   std::unique_ptr<InstructionSelector> InstSelector;
   std::unique_ptr<LegalizerInfo> Legalizer;
@@ -110,7 +113,12 @@ class SBFSubtarget : public SBFGenSubtargetInfo {
   // subtarget options.  Definition of function is auto generated by tblgen.
   void ParseSubtargetFeatures(StringRef CPU, StringRef TuneCPU, StringRef FS);
   bool getHasAlu32() const { return HasAlu32; }
-  bool getHasDynamicFrames() const { return HasDynamicFrames; }
+  bool getHasDynamicFrames() const {
+    return HasDynamicFrames || HasDynamicFramesV3;
+  }
+  bool isDynamicFramesV1() const {
+    return HasDynamicFrames;
+  }
   bool getUseDwarfRIS() const { return UseDwarfRIS; }
   bool getDisableNeg() const { return DisableNeg; }
   bool getReverseSubImm() const { return ReverseSubImm; }
diff --git a/llvm/lib/Target/SBF/SBFTargetFeatures.td b/llvm/lib/Target/SBF/SBFTargetFeatures.td
@@ -55,6 +55,9 @@ def FeatureAbiV2 : SubtargetFeature<"abi-v2", "IsAbiV2",
 def FeatureJmp32 : SubtargetFeature<"jmp-32", "HasJmp32", "true",
                                                 "Enable the JMP32 instruction class">;
 
+def FeatureDynamicFramesV3 : SubtargetFeature<"dynamic-frames-v3", "HasDynamicFramesV3", "true",
+                                            "Enable dynamic frames in SBPFv3">;
+
 class Proc<string Name, list<SubtargetFeature> Features>
  : Processor<Name, NoItineraries, Features>;
 
diff --git a/llvm/test/CodeGen/SBF/dynamic_stack_frame_add_and_sub.ll b/llvm/test/CodeGen/SBF/dynamic_stack_frame_add_and_sub.ll
@@ -1,4 +1,5 @@
 ; RUN: llc < %s -march=sbf -mattr=+dynamic-frames | FileCheck %s
+; RUN: llc -O3 -march=sbf -mattr=+dynamic-frames-v3,+alu32 < %s | FileCheck --check-prefix=CHECK-V3 %s
 ;
 ; Source:
 ; int test_func(int * vec, int idx) {
@@ -13,7 +14,7 @@
 define i32 @test_func(ptr noundef %vec, i32 noundef %idx) #0 {
 ; CHECK-LABEL: test_func:
 ; CHECK: add64 r10, -128
-; CHECK-NOT: add64 r10, 128
+; CHECK-V3: add64 r10, 128
 entry:
   %vec.addr = alloca ptr, align 8
   %idx.addr = alloca i512, align 4
diff --git a/llvm/test/CodeGen/SBF/many_args_new_conv.ll b/llvm/test/CodeGen/SBF/many_args_new_conv.ll
@@ -1,7 +1,7 @@
 ; RUN: llc -O2 -march=sbf -mcpu=v1 < %s | FileCheck %s
 ; RUN: llc -O2 -mtriple=sbpfv1-solana-solana < %s | FileCheck %s
 ; RUN: llc -O2 -march=sbf -mcpu=v1 -mattr=+mem-encoding < %s | FileCheck %s
-; RUN: llc -O3 -march=sbf -mcpu=v3 < %s | FileCheck --check-prefix=CHECK-V3 %s
+; RUN: llc -O3 -march=sbf -mattr=+dynamic-frames-v3 < %s | FileCheck --check-prefix=CHECK-V3 %s
 
 ; Function Attrs: nounwind uwtable
 define i32 @caller_no_alloca(i32 %a, i32 %b, i32 %c) #0 {
@@ -10,15 +10,20 @@ entry:
 
 ; No changes to the stack pointer
 ; CHECK-NOT: add64 r10
-; Add zero to stack pointer from V3 onwards
-; CHECK-V3: add64 r10, 0
 
 ; Saving arguments on the stack
 ; CHECK: stdw [r10 - 40], 60
 ; CHECK: stdw [r10 - 32], 55
 ; CHECK: stdw [r10 - 24], 50
 ; CHECK: stdw [r10 - 16], 4
 ; CHECK: stdw [r10 - 8], 3
+
+; CHECK-V3: stdw [r10 + 40], 60
+; CHECK-V3: stdw [r10 + 32], 55
+; CHECK-V3: stdw [r10 + 24], 50
+; CHECK-V3: stdw [r10 + 16], 4
+; CHECK-V3: stdw [r10 + 8], 3
+
 ; CHECK: mov64 r4, 1
 ; CHECK: mov64 r5, 2
 ; CHECK: call callee_alloca
@@ -45,6 +50,18 @@ define i32 @caller_alloca(i32 %a, i32 %b, i32 %c) #0 {
 ; CHECK: stdw [r10 - 16], 4
 ; Offset in the callee: frame_size - 8
 ; CHECK: stdw [r10 - 8], 3
+
+; Offset in the callee: frame_size - 40
+; CHECK-V3: stdw [r10 + 40], 60
+; Offset in the callee: frame_size - 32
+; CHECK-V3: stdw [r10 + 32], 55
+; Offset in the callee: frame_size - 24
+; CHECK-V3: stdw [r10 + 24], 50
+; Offset in the callee: frame_size - 16
+; CHECK-V3: stdw [r10 + 16], 4
+; Offset in the callee: frame_size - 8
+; CHECK-V3: stdw [r10 + 8], 3
+
 ; CHECK: mov64 r4, 1
 ; CHECK: mov64 r5, 2
 ; CHECK: call callee_no_alloca
@@ -63,6 +80,8 @@ entry:
 define i32 @callee_alloca(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e, i32 %f, i32 %p, i32 %y, i32 %a1, i32 %a2) #1 {
 ; CHECK-LABEL: callee_alloca
 ; CHECK: add64 r10, -128
+; CHECK-V3: add64 r10, 128
+
 ; Loading arguments
 ; CHECK: ldxw r2, [r10 + 120]
 ; CHECK: ldxw r2, [r10 + 112]
@@ -71,6 +90,16 @@ define i32 @callee_alloca(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e, i32 %f, i32 %p
 ; CHECK: ldxw r2, [r10 + 88]
 ; Loading allocated i32
 ; CHECK: ldxw r0, [r10 + 24]
+
+; CHECK-V3: ldxw r2, [r10 - 120]
+; CHECK-V3: ldxw r2, [r10 - 112]
+; CHECK-V3: ldxw r2, [r10 - 104]
+; CHECK-V3: ldxw r2, [r10 - 96]
+; CHECK-V3: ldxw r2, [r10 - 88]
+; Loading allocated i32
+; CHECK-V3: ldxw r0, [r10 - 24]
+
+
 ; CHECK-NOT: add64 r10, 128
 
 entry:
@@ -93,13 +122,22 @@ entry:
 define i32 @callee_no_alloca(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e, i32 %f, i32 %p, i32 %y, i32 %a1, i32 %a2) #1 {
 ; CHECK-LABEL: callee_no_alloca
 ; CHECK: add64 r10, -64
+; CHECK-V3: add64 r10, 64
+
 ; Loading arguments
 ; CHECK: ldxw r1, [r10 + 56]
 ; CHECK: ldxw r1, [r10 + 48]
 ; CHECK: ldxw r1, [r10 + 40]
 ; CHECK: ldxw r1, [r10 + 32]
 ; CHECK: ldxw r1, [r10 + 24]
 
+; Loading arguments
+; CHECK-V3: ldxw r1, [r10 - 56]
+; CHECK-V3: ldxw r1, [r10 - 48]
+; CHECK-V3: ldxw r1, [r10 - 40]
+; CHECK-V3: ldxw r1, [r10 - 32]
+; CHECK-V3: ldxw r1, [r10 - 24]
+
 ; CHECK-NOT: add64 r10, 64
 entry:
   %g = add i32 %a, %b
diff --git a/llvm/test/CodeGen/SBF/many_args_value_size.ll b/llvm/test/CodeGen/SBF/many_args_value_size.ll
@@ -1,5 +1,6 @@
 ; RUN: llc -march=sbf -mcpu=v2 < %s | FileCheck %s
 ; RUN: llc -mtriple=sbpfv2-solana-solana < %s | FileCheck %s
+; RUN: llc -O3 -march=sbf -mattr=+dynamic-frames-v3,+alu32 < %s | FileCheck --check-prefix=CHECK-V3 %s
 
 define i64 @test_func(i64 %a, i64 %b, i64 %c, i64 %d, i64 %e) {
 start:
@@ -10,6 +11,11 @@ start:
 ; CHECK: stw [r10 - 12], 65516
 ; CHECK: stw [r10 - 4], 5
 
+; CHECK-V3: stdw [r10 + 32], 5400
+; CHECK-V3: stw [r10 + 20], 300
+; CHECK-V3: stw [r10 + 12], 65516
+; CHECK-V3: stw [r10 + 4], 5
+
   %res = call i64 @func(i64 %a, i64 %b, i64 %c, i64 %d, i64 %e, i8 5, i16 -20, i32 300, i64 5400)
   ret i64 %res
 }
@@ -18,26 +24,31 @@ define i64 @func(i64 %a, i64 %b, i64 %c, i64 %d, i64 %e, i8 %b8, i16 %b16, i32 %
 start:
 ; CHECK-LABEL: func:
 ; CHECK: add64 r10, -64
+; CHECK-V3: add64 r10, 64
   %a1 = add i64 %a, %b
   %a2 = sub i64 %a1, %c
   %a3 = mul i64 %a2, %d
   %a4 = add i64 %a3, %e
 
 ; -64 + 32 = -32, so this is 5400 in %a5
 ; CHECK: ldxdw r4, [r10 + 32]
+; CHECK-V3: ldxdw r4, [r10 - 32]
 
 ; -64 + 60 = -4, so this is 5 in %b8
 ; CHECK: ldxb w4, [r10 + 60]
+; CHECK-V3: ldxb w4, [r10 - 60]
   %c0 = trunc i64 %a to i8
   %b1 = add i8 %b8, %c0
 
 ; -64 + 52 = -12, so this is -20 in %b16
-; ldxw w1, [r10 + 52]
+; CHECK: ldxh w1, [r10 + 52]
+; CHECK-V3: ldxh w1, [r10 - 52]
   %c1 = trunc i64 %b to i16
   %b2 = add i16 %b16, %c1
 
 ; -64 + 44 = -20, so this is 300 in %b32
 ; CHECK: ldxw w1, [r10 + 44]
+; CHECK-V3: ldxw w1, [r10 - 44]
   %c2 = trunc i64 %c to i32
   %b3 = add i32 %b32, %c2
 

Original file line number	Diff line number	Diff line change
`@@ -57,6 +57,7 @@ void SBFSubtarget::initializeEnvironment(const Triple &TT) {`
`57`	`57`	`HasStaticSyscalls = false;`
`58`	`58`	`IsAbiV2 = false;`
`59`	`59`	`HasJmp32 = false;`
	`60`	`+ HasDynamicFramesV3 = false;`
`60`	`61`	`}`
`61`	`62`
`62`	`63`	`void SBFSubtarget::initSubtargetFeatures(StringRef CPU, StringRef FS) {`